Domain-invariant Face Recognition using Learned Low-rank Transformation
This work addresses domain-invariant face recognition, which is crucial for improving accuracy in real-world applications, though it appears incremental as it builds on existing transformation methods.
The paper tackles the problem of face recognition across varying visual domains like pose and illumination by learning a low-rank transformation that reduces intra-class variations and increases inter-class separations, achieving demonstrated effectiveness in extensive experiments on public datasets.
We present a low-rank transformation approach to compensate for face variations due to changes in visual domains, such as pose and illumination. The key idea is to learn discriminative linear transformations for face images using matrix rank as the optimization criteria. The learned linear transformations restore a shared low-rank structure for faces from the same subject, and, at the same time, force a high-rank structure for faces from different subjects. In this way, among the transformed faces, we reduce variations caused by domain changes within the classes, and increase separations between the classes for better face recognition across domains. Extensive experiments using public datasets are presented to demonstrate the effectiveness of our approach for face recognition across domains. The potential of the approach for feature extraction in generic object recognition and coded aperture design are discussed as well.